详情

原文传递 Winter Road Condition Recognition Using Video Image Classification.

题名：	Winter Road Condition Recognition Using Video Image Classification.
作者：	Kuehnle-A; Burghout-W
关键词：	Winter maintenance; Video cameras; Image processing; Decision making; Neural networks; Accuracy; Classification; Sweden; Road conditions; Artificial neural networks
摘要：	Sweden spends 1.7 billion Crowns (approximately 200 million U.S. dollars) on winter road maintenance annually. A large part of this money goes into plowing, salting, and sanding of the roads. The decision about what maintenance to perform is made, in part, based on data received from road weather information stations, some of which are also equipped with video cameras. These video cameras form an additional unexploited sensor for determining the road condition during winter. Images taken from a handheld roadside video camera are investigated here to see if it is possible to determine the road state (dry, wet, snowy, icy, snowy with tracks) from the video images alone. The system is intended to supplement the other weather station measurements, such as temperature and wind speed, and make better maintenance decisions and quality control of maintenance possible. The results indicate that it is possible to distinguish between all road states except for ice/wet and ice/tracks. Typical class separations are a Mahanalobis distance between 0 and 2. Neural networks with three or four input features, three to five hidden neurons, and a sigmoid-sigmoid-linear architecture are used to classify the road state. Rates of correct classification are typically 40% to 50% with these networks. There are useful feature combinations, including purely monochrome features, which do not depend on the network architecture.
总页数：	Transportation Research Record. 1998. (1627) pp 29-33 (FIGS: 2 Fig. TABS: 2 Tab. PHOT: 1 Phot. REFS: 7 Ref. )
报告类型：	科技报告